Leveraging Generative AI to Design Therapeutic Antibodies

Episode Summary

Executive Summary: James Field says LabGenius uses AI, automation, and synthetic biology to turn antibody discovery into a closed-loop, data-driven process. By generating proprietary ML-grade functional assay data, the company claims it can explore far larger design spaces, co-optimize multiple properties, and find highly selective antibodies—especially for hard targets like solid tumors with on-target/off-tumor toxicity.

Main Topics: Why antibodies are a strong AI target (Priority: 5/5): Field argues antibodies are unusually programmable, modular, and amenable to synthetic biology, making them better suited than many modalities for machine learning-guided design. LabGenius’ differentiated AI discovery platform (Priority: 5/5): The company focuses on disease-relevant, functional cell-based assays rather than structure or binding prediction, using automation and software to generate bespoke training data. Closed-loop active learning for lead optimization (Priority: 5/5): LabGenius runs iterative design-build-test-learn cycles to search large combinatorial spaces and improve molecules across multiple properties at once. Target selection and lead program strategy (Priority: 4/5): The company is prioritizing well-validated targets in solid tumors where protein engineering can solve on-target/off-tumor toxicity, reducing biological risk. Programmable antibodies and future logic-based therapies (Priority: 4/5): Field envisions antibodies that can encode logic, distinguishing healthy from diseased cells and triggering therapeutic actions conditionally. Business model, partnerships, and financing (Priority: 3/5): LabGenius is pursuing a hybrid strategy: advancing an internal pipeline while partnering to expand the platform and generate reusable data. Skepticism about AI in drug discovery (Priority: 3/5): Field acknowledges legitimate skepticism and says AI will prove itself by solving genuinely hard, high-value problems rather than being applied everywhere.

Key Arguments: Antibodies are ideal for AI-driven design because they are programmable, modular, and can be recombined into complex functional architectures. Traditional discovery samples too little of the possible design space; AI lets LabGenius search much larger regions systematically. LabGenius’ edge is proprietary, high-quality functional data from disease-relevant cell assays, not public datasets or simple binding measurements. The platform is designed to co-optimize many properties in parallel, avoiding the classic tradeoff where improving one attribute worsens another. Human intuition can bias design; reducing those biases allowed the system to generate non-obvious, high-performing molecules. The main value of AI in this setting is not just speed or cost reduction, but better molecules with higher chances of clinical success. Partnerships are only valuable if they expand the platform, open new modalities, or create reusable data for future programs. Skepticism toward AI is justified when it is applied broadly without clear fit; the field will mature as the right use cases succeed.

Data Points: Design space explored in one example: Over 250,000 T cell engager designs - Field uses this as an example of the combinatorial search space LabGenius can model and explore. Typical comparison set at major biotech/pharma: 500 to 1,000 molecules - He contrasts this with the much smaller number of variants traditionally tested in a program. Improvement in killing selectivity: At least 400 times better than clinical benchmarks - LabGenius says its platform found molecules with killing selectivities far beyond existing clinical references. Data needed per cycle: A couple of hundred designs characterized per cycle - Field says model training typically requires substantial experimental data each round.

Pivotal Quotes: "The molecule with that right blend of properties exists. It's just a question of how do you find it?" — James Field: Explaining LabGenius’ view of antibody discovery as a search problem across design space. "The process and the approach that we're taking in the application of machine learning to antibody engineering gives you molecules that you just wouldn't have found using conventional methods." — James Field: Describing how AI changes discovery beyond simple speed or cost advantages. "AI and machine learning, it's just a tool and it's not always the best tool." — James Field: On industry skepticism and the need to apply AI only where it offers clear value.

Implications: AI in biologics will matter most where it creates proprietary data and solves hard optimization problems. If LabGenius is right, the field will shift from artisanal antibody discovery toward systematic, programmable therapeutic engineering.